Functional Neural Networks Model for Prediction of the Formation Tops in Real-Time While Drilling

Abstract

Abstract The determination of formation tops while drilling plays a pivotal role in the efficiency and cost-effectiveness of the drilling operations. Identifying lithology changes in real-time is crucial for adapting drilling programs, optimizing well designs, and ensuring the overall success of the drilling process. Real-time detection of lithology changes provides a valuable tool for mitigating uncertainties associated with geological data limitations, especially during the exploration phase. As formations vary in composition and characteristics, the ability to predict these changes enhances the overall management of drilling operations, minimizing risks and contributing to the economic viability of oil well projects. Current methods for detection of the formation tops rely on geological data, introducing uncertainties, especially in exploration due to data limitations. This study explores the real-time predictive capabilities of the functional neural networks (FNNs) for the prediction of the formation tops. Trained on 3162 datasets of six drilling parameters, the FNNs model aims to predict lithology changes and formation tops across the sandstone, anhydrite, carbonate with shale streaks, and carbonate formations. Testing on 1356 datasets from a different well validated the FNNs model. Results affirm the FNNs accurately predicted the carbonate/shale formation top in training data, while it struggled to accurately predict tops for all formations in testing data compared to the reported high accuracy for the artificial neural networks model.